See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/372057110

Diabetes Prediction using Machine Learning Techniques

Article in Journal of Artificial Intelligence and Capsule Networks · June 2023

DOI: 10.36548/jaicn.2023.2.008

CITATIONS READS

8 366

5 authors, including:

V. Jithendra
Madanapalle Institute of Technology & Science
1 PUBLICATION 8 CITATIONS

SEE PROFILE

All content following this page was uploaded by V. Jithendra on 14 July 2023.

The user has requested enhancement of the downloaded file.

Journal of Artificial Intelligence and Capsule Networks (ISSN: 2582-2012)
www.irojournals.com/aicn/

Diabetes Prediction using Machine

Learning Techniques
V Jithendra1, R M Sai Mohit2, M Madhusudhan3, B Jagadeesh4,
Dr. S Kusuma5
1,2,3,4
Student, Dept. Of CSE, MITS, Madanapalle, Andhra Pradesh, India
5
Assistant Professor, Dept. Of CSE, MITS, Madanapalle, Andhra Pradesh, India

Email: 119699A0520@mits.ac.in, 219699A0544@mits.ac.in, 319699A0524@mits.ac.in, 419699A0517@mits.ac.in,

5kusumas@mits.ac.in

Abstract

Now a day due to hectic schedules and sedentary lifestyle people do not follow the
proper diet. Poor diet may lead to diabetes, and which could result in various health issues such
as heart attacks, strokes, renal failure, nerve damage, etc. When diabetes is accurately detected
in its early stage, it can be effectively treated. By using Machine Learning methods, the problem
can be easily detected and a solution could bearrived. Early diabetes detection and prediction
can be greatly improved with machine learning (ML) approaches. When it is detected in an
early stage, it can be resolved quickly. The objective of this research is to provide prediction
using various supervised machine learning methods. Seven algorithms are compared with each
other to figure out which is the best. The algorithms are Logistic Regression, Random Forest,
Decision Tree, K-Nearest Neighbor, Support Vector Machine, Naïve Bayes, Gradient
Boosting. The evaluation results stated that Logistic Regression is more accurate than other
algorithms for the given data set with an accuracy of 82%. After selecting the ML model which
is more accurate. A User Interface where users can enter the new data and get results was
developed and the results to the user were forwarded through WhatsApp along with some
suggestions and precautions.

Keywords: Logistic Regression, Random Forest, Decision Tree, K-Nearest Neighbor, Support
Vector Machine, Naïve Bayes, Gradient Boosting, Machine Learning (ML), Diabetes.

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2, Pages 190-206 190
DOI: https://doi.org/10.36548/jaicn.2023.2.008
Received: 03.05.2023, received in revised form: 03.06.2023, accepted: 18.06.2023, published: 28.06.2023
© 2022 Inventive Research Organization. This is an open access article under the Creative Commons Attribution-NonCommercial International (CC BY-NC 4.0) License
 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

1. Introduction

Diabetes affects numerous people as a result of its daily increase. Most

patients are unaware of their health situation and the challenge before a disease is
diagnosed. The main challenge is to make this architecture flexible across several
datasets and to resolve or improve the prediction model's correctness. The
International Diabetes Federation's seventh edition of the Diabetes Atlas contains
the most recent data on DM (IDF). The number of diabetic patients worldwide in
2015 was over 410 million.

To lessen the impact of diabetes and manage the condition, one must focus
on a person who is at high risk. The World Health Organization (WHO) specifies
additional risk factors for diabetes mellitus as follows:

• Age group above 45 who follows the sedentary lifestyle.

• BMI is more than 24 kg/m2.

• Blood sugar levels during fasting are continuously above the normal
range or blood glucose is elevated above normal levels (IFG).

• Pregnant women whose age is above 30.

• Having high blood pressure.

• Having high blood levels of triglycerides.

• Having a family history with diabetes.

Diabetes is a condition when there is insufficient insulin in the blood, which

leads to deficiencies. Frequent urine and a thirsty are symptoms of elevated blood
sugar. If not treated well , it can cause serious problems or even death .
Cardiovascular disease, an athlete's foot, and blurry vision are also symptoms of
severe difficulty. High blood sugar levels are a sign of prior diabetes. Prior diabetes
does not significantly outweigh traditional value. The exocrine gland's failure to
produce enough hypoglycaemic agents and its improper response to those agents is
what causes diabetes.

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 191
Diabetes Prediction Using Machine Learning Techniques

To carry out the experiment, a dataset of a patient’s medical records is

acquired, and various classification methods are then applied it. The efficiency and
accuracy of the employed algorithms are investigated and compared. The algorithm
that is most effective for prediction will be selected among the different algorithms
that are used. This study uses machine learning approaches to help medical
professionals identify and treat diabetes early.

2. Literature Survey

For a prediction they performed 5 different supervised learning algorithms

on the dataset that was collected [1]. Additionally, the second author used a heat
map to determine the relationship among the features. They used T-Test to identify
the features after analyzing how all the features relate to each other. They conclude
that Logistic Regression is the most accurate learning technique after testing with
various classification models. They are achieving an accuracy of more than 79%
with this method.

Following thorough testing, they chose to use the logistic regression

technique. It was executed using the fundamental RELU formula, and its behavior
was examined [2]. The other study regarding diabetes prediction highlighted that the
current value is used to predict the unknown value. The expected value only has two
possible outcomes: 0 (No) or 1 (Yes). They used various machine learning methods
to predict this diabetes using existing values.

Glucose levels, blood pressure, insulin, body mass index (BMI), age is
considered when developing the model for diabetes prediction. Following that, a set
of real-world data is used to evaluate the method. They achieved accuracy of 72%
for the decision tree and 76.5% for the random forest [3]. Furthermore, the
researchers used the support Vector Machine (SVM) and Random Forest (RF)
classifiers to predict diabetes. The data set is first pre-processed after being obtained
from a clinic.

Some duplicate and incomplete data are removed during the pre-processing
of the dataset. The features are chosen from the data set after that pre-processing.
The dataset is then prepared for dimensionality reduction. splitted The data set is

ISSN: 2582-2012 192

 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

spilt into two portions,i.e., training and testing. Later trained with SVM classifier
and a Random Forest model, respectively. Two models are then evaluated. the
evaluation results showed 81.4% and 83% accuracy respectively [4].

The dataset is spilt as 70% for training and 30% testing . To predict
diabetes, the seven best machine learning algorithms are being deployed. For this
dataset, random forest achieves maximum accuracy. For the test data set, it provides
an accuracy of 79.9%. In consideration of this, they came to the conclusion that,
Random Forest algorithm is the most effective [5]. They primarily take into account
three data sets to predict diabetes, using the same dataset as the other researchers for
prediction. They are thinking about the datasets for heart disease (HD), liver
disease, and diabetes disease.

The proposed model has 97% prediction accuracy. To determine the highest
accuracy, they have used LR formula, this article used GA to forecast by combining
the effects of a sizable number of independent variables.

The proposed classifier's classification result demonstrates that using a

variety of variables to predict problem development is more accurate than using a
single important factor or a few sets important variables [6]. Researchers have
developed a useful approach to predict diabetes with the help of distributed machine
learning and big data platforms like Spark. Through the distributed machine
learning (ML) built on Apache Spark, their work intends to develop models that can
identify diabetes in this circumstance.

and Comparison analysis of five machine learning algorithms is done using

three performance metrics, the accuracy, recall, and precision [7].

3. Proposed Work

The proposed methods implement the seven-machine learning (ML)

classification models and uses the algorithm which performs well in the Graphical
User Interface among the seven algorithms after performing the comparative
analysis for the given dataset. After finding the highest accuracy algorithm that is
used for testing new data. The user enters the new data in the Graphical User
Interface, and it will predict whether the person is diabetic or non-diabetic.

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 193
Diabetes Prediction Using Machine Learning Techniques

3.1 Methodology

As shown in figure 1 the process will be carried out.

Figure 1. Methodology Representation

3.2 Data Collection

A dataset in machine learning is a group of data points used to develop a

model. The accuracy and efficiency of the final model depend greatly on the
dataset's quality. It is the process of collecting an existing data. the diabetes dataset
from the Kaggle website that contains 768 patient entries like Glucose, Insulin, Skin
Thickness, BMI, an Age, DiabetesPedigreeFunction and Outcome was used in the
work to perform the analysis.

ISSN: 2582-2012 194

 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

3.3 Data Pre-processing and Data Analysis

3.3.1 Data Pre-Processing

It is the most important and key aspect in this project. Missing values and
other unwanted values will might reduce the effectiveness of the results. Data pre-
processing is performed to increase the accuracy and effectiveness of the results.
The two step Pre-processing followed in the work is as follows .

1) Missing Values Removal: Remove any value that have null or 0. There
can never be a value of zero. This instance is therefore no longer valid. In order to
reduce the dimensionality of the data and work more quickly, feature subsets are
created by eliminating pointless features and occurrences.

2) Splitting Of Dataset: Splitting of data help us to know how the model will
perform for the new data. The dataset is spilt into two portions each 80% and 20%.
80% is for training and 20% is for testing respectively. This process will be done
after the above step.

3.3.2 Feature Selection

Performance of a Machine learning models can be improved with the help of

a feature selection process. Heat map helps us to know how each and every feature
are correlated and helps us to find the most important features that should be
included in the model building. In this process four features from the original set of
features using heat map for the model building is chosen they are Glucose, BMI,
Insulin and Diabetes Pedigree Function. Figure 2 is used for feature selection.

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 195
Diabetes Prediction Using Machine Learning Techniques

Figure 2. Heat Map

3.4 Model Construction and Prediction

3.4.1 Simulation Tool/ Framework

The whole project is done in the “Jupyter Notebook”. It is an open-source

web application which provides an interactive, convenient environment for
development and documentation of a python project.

3.4.2 Classification Algorithms

a) Logistic Regression

It is a statistical method used to analyse data and make predictions about the
likelihood of an event occurring. When the dependent variable is categorical or
binary, this kind of regression analysis is employed. Figuring out how the variable
are related to each other is the aim of logistic regression. It uses a logistic function
to transform the input values into a probability value between 0 and 1. The logistic
regression model calculates the independent variable coefficients that increase the
probability of the observed data. The chance that the dependent variable will have a
particular value is determined using these coefficients.

ISSN: 2582-2012 196

 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

b) K- Nearest Neighbor

It is a non-parametric algorithm. It operates by determining the K-nearest

points, based on a selected distance metric, to a particular data point. Regression
uses the K-nearest neighbor to calculate the predicted number. It is simple to
implement and works well in low-dimensional spaces. However, it can be
computationally expensive in high-dimensional spaces and requires enough labelled
data. The choice of K affects the accuracy of the model, with small values leading to
more flexible models and larger values leading to more rigid models. KNN can also
suffer from the curse of dimensionality, where the data becomes increasingly sparse,
and the distance metric becomes less meaningful in high-dimensional spaces.

c) Support Vector Machine

It is employed for regression and classification analysis. Finding the

hyperplane that divides data into various groups is how it functions. The hyperplane
is selected for increasing of margin. By utilising kernel functions, this algorithm is
able to manage data that can be separated in both linear and non-linear ways. The
use of regularisation parameters allows them to effectively operate in high-
dimensional environments and avoid overfitting.

d) Naive Bayes

A probabilistic method used in the machine learning for classification jobs is

called naive Bayes. The Bayes theorem states that the likelihood of the evidence
given the hypothesis, multiplied by the prior probability of the hypothesis,
determine. The "naive" premise behind Naive Bayes is that, given class label, the
features are conditionally independent. This makes the algorithm more
computationally effective and simplifies the likelihood estimate.

e) Decision Tree

In this Internal node represent tests on a feature. Each branch represents the
test's result, and each leaf node represents a class name or a numerical value in this
tree-like structure. Because they are simple to manage both categorical and
numerical data.

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 197
Diabetes Prediction Using Machine Learning Techniques

f) Random Forest

It is constructed with the help of multiple decision trees and takes one final
prediction. Each decision tree is grown using a random subset of features and data
samples, making them less prone to overfitting. The final prediction of the Random
Forest is the majority vote or an average prediction of all the individual trees. This
is most frequently used in classification and regression tasks. This algorithm can
handle both categorical and continuous data, and also handles missing values. It can
identify feature importance, making it useful for feature selection and data
visualization. Random Forest is more efficient and can handle huge datasets.

g) Gradient Boosting

Gradient Boosting that involves in creating an ensemble of weak models that

are trained sequentially to correct the errors of the previous models. Each
subsequent model tries to minimize the residual errors of the previous models using
gradient descent optimization. The prediction is based on a weighted sum of the
predictions of all the weak models in the ensemble. The learning rate that controls
the contribution of each individual model to the final prediction, is an important
hyperparameter in this algorithm. It is known for its ability to handle complex
datasets.

3.5 Performance Metrics

We usually use the different machine learning algorithms, and we require

certain tools to assess how well they did their work and how they perform. There
are many performance metrics that are used in machine learning. In this study, we
collect useful data regarding algorithm performance and conduct a comparative
analysis using several widely used measures for various tasks such as Accuracy,
precision, recall, and f1-score.

Accuracy

It is arguably the most common and basic choice for evaluating an

algorithm's effectiveness in classification problems. It can be defined as the
percentage of data items that were properly identified out of all observations.
Although accuracy has a wide range of applications, it is not necessarily the ideal

ISSN: 2582-2012 198

 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

performance metric, particularly if the target variable classes in the sample are
uneven. Mathematically it can be represented as,

𝐴 = ( 𝑇𝑃 + 𝑇𝑁 ) / ( 𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁 (1)

Precision

It is a metric for how accurate a binary classification algorithm is. It is defined as the
ratio of the model's correct positive predictions to its total positive predictions. Mathematically
it can be represented as:

𝑃 = 𝑇𝑃 / ( 𝑇𝑃 + 𝐹𝑃 ) (2)

Recall

In machine learning, recall is a performance measure that is used to assess performance.

The percentage of actual positive instances in the data set that are true positives (i.e., instances
that were properly identified as positive). Mathematically it can be represented as:

𝑅 = 𝑇𝑃 / ( 𝑇𝑃 + 𝐹𝑁 ) (3)

F1-score

It is also known as fscore or f-measure. It will determine how will an algorithm is

performing based on recall and accuracy. It has the following mathematical representation:

𝐹1 = 2 ∗ ((𝑃 ∗ 𝑅) / (𝑃 + 𝑅)) (4)

3.6 Comparative Analysis

A performance evaluation of various classifiers is performed. We have

found that, “Logistic Regression Algorithm” are achieving the highest accuracy
among all the algorithms. Here table 1 shows the scores of all algorithms and figure
3 is comparative analysis of each algorithm.

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 199
Diabetes Prediction Using Machine Learning Techniques

Table 1. Comparative Analysis of Algorithms

Figure 3. Comparative Analysis Plot

3.7 Graphical User Interface

For a simple and interactive interface for user we built the GUI application
using Tkinter package.

Tkinter: It is a standard Python package for creating graphical user interfaces

(GUIs). It provides a set of tools and widgets that allow developers to build desktop
applications with interactive windows, buttons, menus, and other GUI elements.

ISSN: 2582-2012 200

 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

Creation of Tkinter GUI involves the following steps:

Step 1: Installing the Tkinker package and importing it into python script, which
provides required classes and functions for creating a GUI.

Step 2: Creating the main window and customizing it by modifying the attributes
like the title, size.

Step 3: After Creating the main window, now create and add widgets to it. It
involves widgets like labels, buttons, and entry fields, etc.

Here we created 6 labels and 6 entry fields for the user’s name, phone number
and features and Predict, Report buttons.

Step 4: Now positioning the created widgets using a layout manager (Grid
Manager).

Step 5: Binding functions to widget events or defining event handlers.

After clicking on the predict button it displays the result like “Diabetic” or
“Non-Diabetic” and similarly after clicking on the report button it sends report to
the user’s mobile number through WhatsApp.

Step 6: Finally run the main loop

3.8 Sending Reports through WhatsApp

We are using PyWhatKit package for sending report through WhatsApp.

PyWhatKit: It is a Python library that provides various functionalities related

to WhatsApp automation.

Sending reports using PyWhatKit involves the following steps:

Step 1: Installing the PyWhatKit package and importing it into python script.

Step 2: Use the sendwhatmsg() function to send a WhatsApp message. This

function takes the phone number of the user, the message content (Report), the
hour (in 24-hour format), and the minute when you want to send the message.

syntax: pwk.sendwhatmsg(ph_no, msg, hour, minute)

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 201
Diabetes Prediction Using Machine Learning Techniques

where;

ph_no: The phone number of the user entered in GUI, including the country
code.

msg: Detailed text report.

Hour and minute: These are set for two minutes i.e.00:02 from the point of
time when user clicks the report button.

4. Results

One of the major illnesses that might lead to several consequences is

diabetes. According to all of the previous experiments, we found the accuracy of
Logistic Regression is higher than other algorithms for the taken data set. At most
we achieve 82% accuracy for Logistic Regression. After linking the model to the
GUI application, the user enters new data in the Graphical User Interface, and it will
predict whether the person as diabetic or non-diabetic and sends the Report and
Suggestions to the user through WhatsApp message. Below figures are related to
Diabetes Prediction using GUI. Figure 4 shows the user interface, figure 5 shows
the details entered by the user and in figure 6 when user clicks on predict it shows
diabetic or non-diabetic and user clicks on report it sends message as shown in
figure 7.

Figure 4. Graphical User Interface

ISSN: 2582-2012 202

 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

Figure 5. Prediction for the Given Data

Figure 6. Displaying Output and Sending Report to User

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 203
Diabetes Prediction Using Machine Learning Techniques

Figure 7. Detailed Report was Sent to the Patient Through Whatsapp

5. Conclusion

One of the most important real-world health problems is the detection of

diabetes at an early stage. In this study, various procedures and implementation of
various algorithms are done in building the prediction system which results in the
prediction of diabetes. During this work, the research has used the seven machine
learning classification algorithms. These algorithms are studied and evaluated on
different parameters. And the tasks are done on Diabetes Dataset that are collected
from Kaggle website. This study results determine that system achieves an accuracy
of 82% using Logistic Regression algorithm.

References

[1] Arwatki Chen L, Nurul Amin and Soumen Moulik’s “Diabetes Disease
Prediction Using Machine Learning Algorithms” published in IEEE
EMBS Conference on Biomedical Engineering and Sciences (IECBES) in
2020.

ISSN: 2582-2012 204

 V Jithendra, R M Sai Mohit, M Madhusudhan, B Jagadeesh, S Kusuma

[2] Le, T. M., Vo, T. M., Pham, T. N., & Dao, S. V.T. (2021). A Novel Wrapper
Based Feature Selection for Early Diabetes Prediction Enhanced with a
Metaheuristic. IEEE Access, 9,7869–7884.
doi:10.1109/access.2020.3047942.

[3] P, Anirudh Hebbar; M V, Manoj Kumar; H A, Sanjay (2019). [IEEE 2019 1st
International Conference on Advances in Information Technology (ICAIT)
- Chikmagalur, India(2019.7.25- 2019.7.27)] 2019 1st International
Conference on Advances in InformationTechnology (ICAIT) - DRAP:
Decision Tree and Random Forest Based Classification Modelto Predict
Diabetes. 271–276. doi:10.1109/icait47043.2019.8987277

[4] Sivaranjani, S., Ananya, S., Aravinth, J., & Karthika, R. (2021). Diabetes
Prediction using Machine Learning Algorithms with Feature Selection and
Dimensionality Reduction. 2021 7th International Conference on
Advanced Computing and Communication Systems (ICACCS).
doi:10.1109/icaccs51430.2021.9441935

[5] Barhate, Rahul; Kulkarni, Pradnya (2018). [IEEE 2018 Fourth International
Conference on Computing Communication Control and Automation
(ICCUBEA) - Pune, India (2018.8.16- 2018.8.18)] 2018 Fourth
International Conference on Computing Communication Control and
Automation (ICCUBEA) - Analysis of Classifiers for Prediction of Type II
Diabetes Mellitus.1–6. doi:10.1109/ICCUBEA.2018.8697856

[6] Chaudhuri, A. K., & Das, A. (2020). Variable Selection in Genetic Algorithm
Model with Logistic Regression for Prediction of Progression to Diseases.
2020 IEEE International Conference for Innovation in
Technology(INOCON) doi:10.1109/inocon50539.2020.9298372

[7] Ahmed, Hager; Younis, Eman M.G.; Ali, Abdelmgeid A. (2020). [IEEE 2020
International Conference on Innovative Trends in Communication and
Computer Engineering (ITCE) - Aswan, Egypt (2020.2.8-2020.2.9)]
2020 International Conference on Innovative Trends in Communication
and Computer Engineering (ITCE) - Predicting Diabetes using Distributed

Journal of Artificial Intelligence and Capsule Networks, June 2023, Volume 5, Issue 2 205
 Diabetes Prediction Using Machine Learning Techniques

Machine Learning based on ApacheSpark*,44–49.

doi:10.1109/ITCE48509.2020.9047795

[8] Debadri Dutta, Debpriyo Paul, Parthajeet Ghosh, "Analyzing Feature

Importance’s for Diabetes Prediction using Machine Learning". IEEE,
pp 942-928, 2018.

[9] K.VijiyaKumar, B.Lavanya, I.Nirmala, S.Sofia Caroline, "Random Forest

Algorithm for the Prediction of Diabetes ".Proceeding of International
Conference on Systems Computation Automation and Networking,
2019.

[10] Md. Faisal Faruque, Asaduzzaman, Iqbal H. Sarker, "Performance Analysis

of Machine Learning Techniques to Predict Diabetes Mellitus".
International Conference on Electrical, Computer and Communication
Engineering (ECCE), 7-9 February, 2019.

[11] Tejas N. Joshi, Prof. Pramila M. Chawan, "Diabetes Prediction Using

Machine Learning Techniques”. Int. Journal of Engineering Research and
Application, Vol. 8, Issue 1, (Part -II) January 2018, pp.-09-13

[12] Nonso Nnamoko, Abir Hussain, David England, "Predicting Diabetes Onset:
an Ensemble Supervised Learning Approach ". IEEE Congress on
Evolutionary Computation (CEC), 2018.

[13] Deeraj Shetty, Kishor Rit, Sohail Shaikh, Nikita Patil, "Diabetes Disease
Prediction Using Data Mining ".International Conference on
Innovations in Information, Embedded and Communication Systems
(ICIIECS), 2017.

[14] Nahla B., Andrew et al,"Intelligible support vector machines for diagnosis of
diabetes mellitus. Information Technology in Biomedicine", IEEE
Transactions. 14, (July. 2010), 1114-20.

[15] A.K., Dewangan, and P., Agrawal, “Classification of Diabetes Mellitus

Using Machine Learning Techniques,” International Journal of
Engineering and Applied Sciences, vol. 2, 2015.

ISSN: 2582-2012 206

View publication stats